Deep Reinforcement Learning Based Optimization Algorithm for Electromagnetic Relays

With the gradual development of electromagnetic relays in the direction of high performance, their structural parameters and performance indexes have become more and more complex. The optimization process of electromagnetic systems often requires repeated calculations of their models to find the optimal structural parameters to achieve higher performance indicators. At present, the more common calculation method is to rely on electromagnetic simulation software and orthogonal design or local optimization algorithms to complete the overall electromagnetic system optimization design process. However, the problems solved by the above methods are limited, and there are problems with slow calculation speed and the tendency to fall into local optimization. Therefore, it is challenging to accurately find the global optimum parameters and quickly complete the optimization design. To address this problem, this paper proposes a deep reinforcement learning-based optimization algorithm for electromagnetic relays. Firstly, by combining migration learning and deep learning, the electromagnetic suction force of the electromagnetic relay under different structural parameters is predicted, which lays a data foundation for the subsequent parameter optimization; secondly, the control strategy of the reinforcement learning algorithm is combined to quickly and accurately find the optimal design strategy for the structural parameters of the electromagnetic system; and lastly, the optimization results based on the above method are verified in simulation experiments, which show that this strategy can be used by itself to find the optimal design. Finally, the optimization results based on the above method are verified in the simulation test, and the test results show that the strategy can effectively search the global optimal parameter strategy of the electromagnetic relay by itself, which improves the performance index of the electromagnetic relay, avoids local fitting, and has a better parameter optimization effect.